【摘要】 随着深度学习技术的飞速发展，AI伪造图像技术日益成熟，对电子取证工作带来了严峻挑战。为应对这一威胁，在深入调研深度伪造技术和图像检测理论的基础上，提出基于ResNeXt网络的深度伪造人脸图像检测模型，并将其应用于电子取证领域。深入剖析了ResNeXt网络在特征提取方面的优势。针对深度伪造人脸图像的独特性，设计了一个多尺度特征融合模块，以有效捕捉伪造图像中细微的伪造痕迹。通过在实际电子取证场景中的应用验证，该方法在FaceDB数据集上取得了显著效果，准确率为88%,曲线下面积(Area Under Curve, AUC)达到0.89,为电子取证提供了强有力的技术支持。更多还原

【Abstract】 With the rapid advancement of deep learning technology, AI-generated image forgery techniques have become increasingly sophisticated, posing significant challenges to digital forensics.In order to address this threat, a ResNeXt-based deepfake facial image detection model is proposed and applied to digital forensics, on the basis of an in-depth investigation of deepfake technologies and image detection theories.First, the advantages of ResNeXt network in feature extraction are thoroughly analyzed.Second, considering the unique characteristics of deepfake facial images, a multi-scale feature fusion module is designed to effectively capture subtle traces in forgery images.Finally, through application verification in real-world digital forensic scenarios, remarkable results are achieved on the FaceDB dataset, with the accuracy of 88% and Area Under Curve(AUC)of 0.89,providing robust technical support for digital forensics.更多还原